In [1]:
import os
import csv
from sklearn.model_selection import train_test_split
import cv2
import numpy as np
import sklearn
from sklearn.utils import shuffle
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import matplotlib
from skimage import exposure
from keras.models import Sequential, Model
from keras.layers.core import Dense, Activation, Flatten, Dropout
from keras.layers.convolutional import Convolution2D, Cropping2D
from keras.layers.pooling import MaxPooling2D
from keras.applications.resnet50 import ResNet50, preprocess_input
from keras.applications.inception_v3 import InceptionV3
from keras.applications.vgg16 import VGG16
from keras.layers import Dense, Flatten, Input, AveragePooling2D, GlobalAveragePooling2D, Lambda, Dropout
from keras.layers.normalization import BatchNormalization
from keras.models import load_model
import keras
from keras.optimizers import Adam
In [2]:
samples = []
b_size = 32
ch, row, col = 3, 160, 320 # Trimmed image format
steering_limit = 1
for root, dirs, files in os.walk("./../../../"):
for file in files:
if file.endswith(".csv"):
#print(os.path.join(root, file))
with open(os.path.join(root, file)) as csvfile:
reader = csv.reader(csvfile)
for line in reader:
# steering_angle = line[3]
# if (steering_angle > 0 and steering_angle < steering_limit):
# samples.append(line)
# elif (steering_angle < 0 and steering_angle > steering_limit):
# samples.append(line)
samples.append(line)
In [3]:
def add_random_shadow(image):
top_y = 320*np.random.uniform()
top_x = 0
bot_x = 160
bot_y = 320*np.random.uniform()
image_hls = cv2.cvtColor(image,cv2.COLOR_RGB2HLS)
shadow_mask = 0*image_hls[:,:,1]
X_m = np.mgrid[0:image.shape[0],0:image.shape[1]][0]
Y_m = np.mgrid[0:image.shape[0],0:image.shape[1]][1]
shadow_mask[((X_m-top_x)*(bot_y-top_y) -(bot_x - top_x)*(Y_m-top_y) >=0)]=1
random_bright = .25+.7*np.random.uniform()
#random_bright = .5
cond1 = shadow_mask==1
cond0 = shadow_mask==0
if np.random.randint(2)==1:
image_hls[:,:,1][cond1] = image_hls[:,:,1][cond1]*random_bright
else:
image_hls[:,:,1][cond0] = image_hls[:,:,1][cond0]*random_bright
image = cv2.cvtColor(image_hls,cv2.COLOR_HLS2RGB)
#image = cv2.cvtColor(image_hls,cv2.COLOR_RGB2HSV)
return image
def trans_image(image,steer,trans_range):
# Translation
tr_x = trans_range*np.random.uniform()-trans_range/2
steer_ang = steer + tr_x/trans_range*2*.2
tr_y = 40*np.random.uniform()-40/2
#tr_y = 0
Trans_M = np.float32([[1,0,tr_x],[0,1,tr_y]])
image_tr = cv2.warpAffine(image,Trans_M,(col,row))
return image_tr,steer_ang
In [4]:
angles_zero_valid = []
angles_left_valid = []
angles_right_valid = []
angles_zero = []
angles_left = []
angles_right = []
train_samples, validation_samples = train_test_split(samples, test_size=0.20)
for tempSample in train_samples:
angle = float(tempSample[3])
if angle == 0:
angles_zero.append(angle)
elif angle > 0:
angles_right.append(angle)
elif angle < 0:
angles_left.append(angle)
for tempSample in validation_samples:
angle = float(tempSample[3])
if angle == 0:
angles_zero_valid.append(angle)
elif angle > 0:
angles_right_valid.append(angle)
elif angle < 0:
angles_left_valid.append(angle)
fig, ax = plt.subplots()
index = np.arange(3)
bar_width = 0.3
opacity = 0.6
rects1 = plt.bar(index, (len(angles_zero),len(angles_right),len(angles_left)), bar_width,
alpha=opacity,
color='b',
label='Train images')
rects2 = plt.bar(index + bar_width, (len(angles_zero_valid),len(angles_right_valid),len(angles_left_valid)), bar_width,
alpha=opacity,
color='r',
label='Test images')
plt.xlabel('Sign ID')
plt.ylabel('Count')
plt.title('Distribution of Dataset')
plt.xticks(index + bar_width / 2, ('0','Left','Right'))
plt.legend()
plt.tight_layout()
plt.show()
This distribution show a pretty balanced ratio between left angles and right angles. However the 0-angle is still to dominant...
In [5]:
img = mpimg.imread(samples[10000][0])
imgplot = plt.imshow(img)
plt.show()
In [6]:
imgplot = plt.imshow(np.fliplr(img))
plt.show()
In [7]:
imgplot = plt.imshow(add_random_shadow(img))
plt.show()
In [8]:
imgplot = plt.imshow(trans_image(img,-0.15,150.0)[0])
plt.show()
In [9]:
model = Sequential()
model.add(Lambda(lambda x: x/255 - .5, input_shape=(row, col , ch)))
model.add(Cropping2D(cropping=((70,25),(0,0))))
model.add(Convolution2D(24,5,5,subsample=(2,2),activation="relu"))
model.add(Dropout(0.1))
model.add(Convolution2D(36,5,5,subsample=(2,2),activation="relu"))
model.add(Convolution2D(48,5,5,subsample=(2,2),activation="relu"))
model.add(Dropout(0.1))
model.add(Convolution2D(64,3,3,activation="relu"))
model.add(Convolution2D(64,3,3,activation="relu"))
model.add(Flatten())
model.add(Dropout(0.1))
model.add(Dense(100))
model.add(BatchNormalization())
model.add(Dense(50))
model.add(BatchNormalization())
model.add(Dense(10))
model.add(BatchNormalization())
model.add(Dense(1))
adam = Adam(lr=1e-4, beta_1=0.9, beta_2=0.999, epsilon=1e-08, decay=0.0)
model.compile(loss='mse', optimizer=adam)
In [10]:
# now show the model summary
model.summary()
In [ ]:
In [38]:
import os.path
recovery = []
for root, dirs, files in os.walk("./../../../training_data_0a/recovery"):
for file in files:
if file.endswith(".jpg"):
abspath = os.path.join('C:\\_Projects\\Udacity\\BehavioralCloning\\training_data_0a\\recovery', file)
recovery.append(abspath)
In [37]:
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import matplotlib
for i in range(len(recovery)):
img = mpimg.imread(recovery[i])
imgplot = plt.imshow(img)
plt.show()
In [ ]:
In [ ]: